Informatic profiling of clinically relevant mutation Many approaches have been developed to predict whether a mutation associated with a disease is actually causative. In contrast to other approaches to predicting deleterious mutations, our approach, called in silico functional profiling, starts with learning residue-specific protein function and then estimates when it is disrupted. This research will continue our efforts to characterize what the underlying molecular disruption a mutation is causing and thereby improve accuracy of these approaches. We will do this by building a database that links clinical observation with molecular phenotype and using this to develop bioinformatic models of mutation. This is particularly relevant to cancer, since many mutations in cancer are both poorly understood and simply associated with cancer. The hypothesis of this proposal is that computational methods that predict a specific residue function using protein sequence and structure can classify known disease-associated mutations based on their function better than existing computational methods, and less expensively than experimental assays. In short, we will describe each phenotypically annotated mutation as possibly affecting catalysis, protein interactions, posttranslational modification and stability of the protein. The structural environments around disease associated mutations can be characterized using a combination of computational biochemical methods based on first principles of biomolecular structure and function and statistical informatics methods. We will continue this research by implementing the following steps: First, we will build a database of how often mutations in cancer, pharmacogenetics, Mendelian and complex disease are disrupted by phosphorylation, stability, catalysis, protein interaction and other posttranslational modifications. Second, we will build a bioinformatic model of disruption using machine learning methods trained with these and other commonly used features. Finally, we will link these to clinical observation by annotating disease causing mutation with an ontology of diseases and integrate these predictions into databases of mutation. Thus, we will link clinical observation with molecular phenotypes by building a useful database and new models of how mutations cause disease.